raw_data <- read_csv("data/wrangled_utterances_representations.csv") 

#WRANGLE
df_insights <- raw_data %>% 
  #rename and factorize columns
  mutate(
    sid = factor(UID), #NOT actually a unique utterance id, treat as sheet order id
    pid = factor(PID),    
    DATASET = factor(recode(Notebook, "Happiness"="happiness", "Space"="space")), #cleanup diff case
    outcomeType = recode(DATASET, "happiness"="numeric", "space"="nominal"),
    top_code = factor(highlevel),
    #recode process 
    top_code = recode(top_code, "ANALYSIS PROCESS" = "PROCESS"),
    top_code = factor(top_code, levels = c("PROCESS","DATASET","VARIABLE","RELATIONSHIP")),
    mid_code = factor(`Data Type`),
    low_code = factor(UtteranceType),
    timestamp = Timestamp,
    repns = group,
    ixn = factor(interaction_used),
    utterance = Utterance,
    TASK = factor(recode(Condition, "Static"="static", "Interactive"="ixn" )),
    uid = factor(as.numeric(factor(paste(pid,factor(utterance))))) #construct a unique ID for utterances
  ) %>% select( #select only needed columns
    sid,uid, pid, TASK, DATASET, ixn, top_code, mid_code, low_code, repns, timestamp, utterance
  )

#DF OF UNIQUE UTTERANCES
df_uniques <- df_insights %>% select(uid, pid, TASK, DATASET) %>% 
  distinct()  #take only unique utterances

glimpse(df_insights)
## Rows: 743
## Columns: 12
## $ sid       <fct> 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17…
## $ uid       <fct> 419, 431, 473, 449, 423, 446, 421, 477, 443, 457, 476, 451, …
## $ pid       <fct> j2719eertu2, j2719eertu2, j2719eertu2, j2719eertu2, j2719eer…
## $ TASK      <fct> static, static, static, static, static, static, static, stat…
## $ DATASET   <fct> space, space, space, space, space, space, space, space, spac…
## $ ixn       <fct> FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALS…
## $ top_code  <fct> DATASET, DATASET, DATASET, VARIABLE, DATASET, VARIABLE, VARI…
## $ mid_code  <fct> NA, NA, NA, distribution (categorical), NA, distribution (ca…
## $ low_code  <fct> "data orientation", "data orientation", "data size", "distri…
## $ repns     <chr> "dataframe", "dataframe", "profile", "profile", "profile", "…
## $ timestamp <time> 19:57:00, 20:10:00, 21:27:00, 21:29:00, 21:38:00, 21:44:00,…
## $ utterance <chr> "\"Alright, so every row is the passenger, their home planet…
glimpse(df_uniques)
## Rows: 662
## Columns: 4
## $ uid     <fct> 419, 431, 473, 449, 423, 446, 421, 477, 443, 457, 476, 451, 48…
## $ pid     <fct> j2719eertu2, j2719eertu2, j2719eertu2, j2719eertu2, j2719eertu…
## $ TASK    <fct> static, static, static, static, static, static, static, static…
## $ DATASET <fct> space, space, space, space, space, space, space, space, space,…

1 DATA PROFILE

TODO TALK WITH DYLAN - resolve missing data in TASK, outcomeType, timestamp… are these the result of ‘exploded’ utterances that were dual coded? <– need to carry the other attributes across both obs - max of 2 detail-codes applied, correct? - where are the flag codes?

df_insights %>% summarytools::dfSummary(
             plain.ascii  = FALSE,
             graph.magnif = 0.75,
             style        = "grid",
             tmp.img.dir  = "temp",
             missing.col = FALSE, 
             method = "render"
)

1.0.1 Data Frame Summary

1.0.1.1 df_insights

Dimensions: 743 x 12
Duplicates: 0

No Variable Stats / Values Freqs (% of Valid) Graph Valid Missing
1 sid
[factor]		 1. 0
2. 1
3. 2
4. 3
5. 4
6. 5
7. 6
8. 7
9. 8
10. 9
[ 733 others ]		 1 ( 0.1%)
1 ( 0.1%)
1 ( 0.1%)
1 ( 0.1%)
1 ( 0.1%)
1 ( 0.1%)
1 ( 0.1%)
1 ( 0.1%)
1 ( 0.1%)
1 ( 0.1%)
733 (98.7%)		 743
(100.0%)		 0
(0.0%)
2 uid
[factor]		 1. 1
2. 2
3. 3
4. 4
5. 5
6. 6
7. 7
8. 8
9. 9
10. 10
[ 652 others ]		 2 ( 0.3%)
1 ( 0.1%)
1 ( 0.1%)
1 ( 0.1%)
1 ( 0.1%)
1 ( 0.1%)
1 ( 0.1%)
2 ( 0.3%)
1 ( 0.1%)
1 ( 0.1%)
731 (98.4%)		 743
(100.0%)		 0
(0.0%)
3 pid
[factor]		 1. 3r2sh20ei
2. 4728sjuiz
3. 7382kwtue
4. 7ACC0B75
5. 8v892iige
6. 92ghd48xe
7. bjs827ee1u
8. E1D39056
9. iurmer289
10. j2719eertu2
[ 3 others ]		 103 (14.0%)
43 ( 5.9%)
54 ( 7.4%)
28 ( 3.8%)
49 ( 6.7%)
56 ( 7.6%)
29 ( 4.0%)
25 ( 3.4%)
87 (11.9%)
78 (10.6%)
182 (24.8%)		 734
(98.8%)		 9
(1.2%)
4 TASK
[factor]		 1. ixn
2. static		 342 (46.6%)
392 (53.4%)		 734
(98.8%)		 9
(1.2%)
5 DATASET
[factor]		 1. happiness
2. space		 420 (57.2%)
314 (42.8%)		 734
(98.8%)		 9
(1.2%)
6 ixn
[factor]		 1. FALSE
2. TRUE		 634 (85.3%)
109 (14.7%)		 743
(100.0%)		 0
(0.0%)
7 top_code
[factor]		 1. PROCESS
2. DATASET
3. VARIABLE
4. RELATIONSHIP		 160 (21.5%)
176 (23.7%)
122 (16.4%)
285 (38.4%)		 743
(100.0%)		 0
(0.0%)
8 mid_code
[factor]		 1. distribution (continuous
2. distribution (categorical
3. relationship (categorical
4. relationship (categorical
5. relationship (continuous
6. relationship (multivariat		 77 (18.0%)
54 (12.6%)
28 ( 6.6%)
55 (12.9%)
146 (34.2%)
67 (15.7%)		 427
(57.5%)		 316
(42.5%)
9 low_code
[factor]		 1. data orientation
2. data provenance
3. data size
4. distribution outlier (var
5. distribution range [min,
6. distribution shape [shape
7. distribution variance (sd
8. missing data
9. outlier (relationship)
10. plan of action
[ 8 others ]		 16 ( 2.2%)
11 ( 1.5%)
9 ( 1.2%)
9 ( 1.2%)
33 ( 4.4%)
79 (10.6%)
1 ( 0.1%)
76 (10.2%)
20 ( 2.7%)
52 ( 7.0%)
437 (58.8%)		 743
(100.0%)		 0
(0.0%)
10 repns
[character]		 1. scatterplot
2. profile
3. none
4. dataframe
5. Multi-view Chart
6. data_dictionary
7. pairplot
8. lineplot
9. describe
10. double-profiler
[ 14 others ]		 128 (17.2%)
107 (14.4%)
105 (14.1%)
74 (10.0%)
59 ( 7.9%)
56 ( 7.5%)
50 ( 6.7%)
36 ( 4.8%)
23 ( 3.1%)
23 ( 3.1%)
82 (11.0%)		 743
(100.0%)		 0
(0.0%)
11 timestamp
[hms, difftime]		 min : 868
med : 70710
max : 215160
units : secs		 622 distinct values 738
(99.3%)		 5
(0.7%)
12 utterance
[character]		 1. [Talking about the profil
2. actually, let me see if p
3. Although we have like les
4. And are they within range
5. And confidence in governm
6. And just I want to see ho
7. And so it looks like it s
8. And then if I had more ti
9. Because it does seem like
10. Data frame. Got a bunch o
[ 652 others ]		 2 ( 0.3%)
2 ( 0.3%)
2 ( 0.3%)
2 ( 0.3%)
2 ( 0.3%)
2 ( 0.3%)
2 ( 0.3%)
2 ( 0.3%)
2 ( 0.3%)
2 ( 0.3%)
723 (97.3%)		 743
(100.0%)		 0
(0.0%)

1.1 DESCRIBE

#COUNTS 
n_rows <- df_insights %>% nrow()
n_unique <- nlevels(df_insights$uid)
n_participants <- nlevels(df_insights$pid)

#count number of codes per unique utterance
s <- df_insights %>% group_by(uid) %>% 
    dplyr::summarise(
      count = n()
) %>% arrange(desc(count), .by_group = TRUE)

max_codes <- max(s$count)

#display frequencies
(f <- freq(s$count,
     order    = "freq",
     rows     = 1:10,
     headings = FALSE))
## There are only 2 rows to show; higher numbers will be ignored
## 
##               Freq   % Valid   % Valid Cum.   % Total   % Total Cum.
## ----------- ------ --------- -------------- --------- --------------
##           1    581     87.76          87.76     87.76          87.76
##           2     81     12.24         100.00     12.24         100.00
##        <NA>      0                               0.00         100.00
##       Total    662    100.00         100.00    100.00         100.00
coded_single <- f[1,1]
coded_double <- f[2,1]

There are 743 coded utterances, representing 662 unique statements made by 13 in the study. 581 utterances were single-coded, while 81 utterances received two detail codes. No more than 2 were applied to any single utterance.

1.2 NUMBER OF UTTERANCES

How many utterances did each participant make in static (vs) interactive tasks?

#SUMMARY DFS
# utterances_by_participant <- df_uniques %>% 
#   group_by(pid, TASK) %>% 
#   dplyr::summarise(
#     n_utterances = n()
#   )


#SUMMARY TABLE
title = "Utterances by Participant and Task"
cols = c("Static Task","Interactive Task","Total Utterances")
cont <- table(df_uniques$pid, df_uniques$TASK)
cont %>% addmargins() %>% kbl(caption = title, col.names = cols) %>%  kable_classic()
Utterances by Participant and Task
Static Task Interactive Task Total Utterances
3r2sh20ei 32 52 84
4728sjuiz 12 27 39
7382kwtue 18 22 40
7ACC0B75 9 17 26
8v892iige 21 24 45
92ghd48xe 28 27 55
bjs827ee1u 15 11 26
E1D39056 14 9 23
iurmer289 39 41 80
j2719eertu2 43 29 72
li832lin23 30 16 46
lkin27js09b 14 23 37
s294hoei 27 55 82
Sum 302 353 655 
# FOR DODGED NOT STACKED
# gf_bar(  pid ~ . , fill = ~ TASK, data = df_insights,
#            position = position_dodge(),
#            orientation = 'y') +

#UTTERANCES by PARTICPANT and TASK (horizontal)
gf_bar(  pid ~ uid , fill = ~ TASK, data = df_uniques) +
# %>%   gf_facet_grid(.~TASK) +
  labs(
    title = "Number of Utterances by Participant and Task",
    subtitle = "some participants were far more talkative than others",
    x = "number of unique utterances",
    y = "participant",
    fill = "Analysis Task"
  )

#SUMMARY TABLE
title = "Utterances by Participant and Dataset"
cols = c("Happiness (numeric)"," Space (nominal)","Total Utterances")
cont <- table(df_uniques$pid, df_uniques$DATASET)
cont %>% addmargins() %>% kbl(caption = title, col.names = cols) %>%  kable_classic()
Utterances by Participant and Dataset
Happiness (numeric) Space (nominal) Total Utterances
3r2sh20ei 52 32 84
4728sjuiz 27 12 39
7382kwtue 18 22 40
7ACC0B75 17 9 26
8v892iige 21 24 45
92ghd48xe 27 28 55
bjs827ee1u 11 15 26
E1D39056 14 9 23
iurmer289 41 39 80
j2719eertu2 43 29 72
li832lin23 30 16 46
lkin27js09b 14 23 37
s294hoei 55 27 82
Sum 370 285 655 
# FOR DODGED NOT STACKED
# gf_bar(  pid ~ . , fill = ~ TASK, data = df_insights,
#            position = position_dodge(),
#            orientation = 'y') +

#UTTERANCES by PARTICPANT and DATASET (horizontal)
gf_bar(  pid ~ uid , fill = ~ DATASET, data = df_uniques) + 
  # %>% gf_facet_grid(.~DATASET) +
  labs(
    title = "Number of Utterances by Participant and Dataset",
    subtitle = "Nominal outcome variable (happiness) tended to yield more utterances",
    x = "number of unique utterances",
    y = "participant",
    fill = "Dataset"
  )

#UTTERANCES by PARTICPANT and DATASET (horizontal)
gf_bar(  pid ~ uid , fill = ~ DATASET, data = df_uniques) %>% 
  gf_facet_grid(.~TASK) + 
  labs(
    title = "Number of Utterances by Participant, Dataset and Task",
    subtitle = "",
    x = "number of unique utterances",
    y = "participant",
    fill = "DATASET"
  )

#UTTERANCES by PARTICPANT, TASK, and DATASET (horizontal)
#FACETED BY PARTICIPANT AND TOP CODE 
gf_bar(  TASK ~ uid , fill = ~ DATASET, data = df_insights) %>%  
  gf_facet_grid(top_code ~ pid) + 
  labs(
    title = "High Level Utterances by Participant, Dataset and Dataset",
    subtitle = "",
    x = "number of utterances",
    y = "Analysis Task",
    fill = "Dataset"
  )

1.3 KINDS OF UTTERANCES

1.3.1 By Task+Dataset

#PREP DATASET FOR VIS
df <- df_insights %>% mutate(
  top_code = fct_rev(top_code), #reorder
  pid = factor(pid, levels = c(
 "bjs827ee1u", "3r2sh20ei", "4728sjuiz", "7ACC0B75","92ghd48xe","iurmer28", "s294hoei", #HAPPINESS-FIRST
  "j2719eertu2","lkin27js09b","li832lin23","7382kwtue","E1D39056","8v892iige")),
  data_order = factor(paste(TASK,"_",DATASET)), #create an order var 
  data_order = recode(data_order, "ixn _ happiness"="space-first",
                                  "ixn _ space"="happiness-first",
                                  "NA _ NA"="NA",
                                  "static _ happiness"="happiness-first",
                                  "static _ space"="space-first")) %>%  
  select(pid,uid,TASK,DATASET,top_code, data_order) %>% 
  #remove handful of rows missing data TODO RESOLVE W/ DYLAN
  na.omit() %>% arrange(data_order)

#SUMMARIZED DF
df <- df_insights %>% mutate(
  top_code = fct_rev(top_code)) %>%  #reorder
  select( TASK, DATASET, top_code) %>% 
  group_by(TASK,DATASET, top_code) %>% 
  dplyr::summarise( 
    c = n()
  )
## `summarise()` has grouped output by 'TASK', 'DATASET'. You can override using
## the `.groups` argument.
##[stacked bar] UTTERANCES by TYPE, TASK W/ COUNTS
ggplot (df, aes(DATASET,c, fill= top_code)) + 
  geom_col() + 
  facet_grid(.~df$TASK)+
  geom_text(aes(label = c), size = 3, hjust = 0.5, vjust = 1.5, position = "stack") +
  scale_fill_brewer(type="qual", palette = 3) +
  theme_minimal()

1.3.2 By Participant

#PREP DATASET FOR VIS
df <- df_insights %>% mutate(
  top_code = fct_rev(top_code), #reorder
  pid = factor(pid, levels = c(
 "bjs827ee1u", "3r2sh20ei", "4728sjuiz", "7ACC0B75","92ghd48xe","iurmer28", "s294hoei", #HAPPINESS-FIRST
  "j2719eertu2","lkin27js09b","li832lin23","7382kwtue","E1D39056","8v892iige")),
  data_order = factor(paste(TASK,"_",DATASET)), #create an order var 
  data_order = recode(data_order, "ixn _ happiness"="space-first",
                                  "ixn _ space"="happiness-first",
                                  "NA _ NA"="NA",
                                  "static _ happiness"="happiness-first",
                                  "static _ space"="space-first")) %>%  
  select(pid,uid,TASK,DATASET,top_code, data_order) %>% 
  #remove handful of rows missing data TODO RESOLVE W/ DYLAN
  na.omit() %>% arrange(data_order)


#TOP CODES BY FACTORS + PARTICIPANTS
(parts <- gf_bar( pid ~ ., fill = ~top_code, data = df) %>% 
  gf_facet_grid(DATASET ~ TASK) + 
  scale_fill_brewer(type="qual", palette = 3)+
  theme_minimal())

##GGPLOT VERSION
ggplot (df) + 
  geom_bar(aes(y = pid, fill = top_code)) + 
  facet_wrap(df$TASK)+
  scale_fill_brewer(type="qual", palette = 3) +
  theme_minimal()

#BIDIRECTIONAL BAR CHART

#SUMMARIZED 
df_summ <- df %>% group_by(pid,TASK,DATASET,top_code,data_order) %>% 
  dplyr::summarise(
    n_utterances = n()
  ) %>% mutate(
    adj_utterances = ifelse(TASK=="static", (n_utterances*-1), (n_utterances*1))
  ) %>% arrange(data_order)
## `summarise()` has grouped output by 'pid', 'TASK', 'DATASET', 'top_code'. You
## can override using the `.groups` argument.
 ggplot (df_summ, aes(x=adj_utterances, y=pid, fill = top_code)) + 
  geom_bar(stat="identity") + 
  facet_wrap(~TASK) +
  scale_fill_brewer(type="qual", palette = 3) +
  theme_minimal()

# df_hap_first <- df %>% filter(data_order == "happiness-first")
# df_space_first <- df %>% filter(data_order == "space-first")
# 
# top <- gf_bar( pid ~ ., fill = ~top_code, data = df_hap_first) %>% 
#   gf_facet_grid(. ~ TASK) + 
#   scale_fill_brewer(type="qual", palette = 3)+
#   theme_minimal()
# 
# bottom <- gf_bar( pid ~ ., fill = ~top_code, data = df_space_first) %>% 
#   gf_facet_grid(. ~ TASK) + 
#   scale_fill_brewer(type="qual", palette = 3)+
#   theme_minimal()
# 
# top+bottom

2 MODELLING

#DEFINE DATAFRAME
df_raw <- df_insights %>% select(pid, uid, TASK, DATASET) %>% mutate(
  TASK = factor(TASK, levels = c("static", "ixn")) #reorder factor levels
) %>% na.omit()
print("WARNING: THE FOLLOWING HAVE OMMITED MISSING DATA RATHER THAN FINDING THE SOURCE")
## [1] "WARNING: THE FOLLOWING HAVE OMMITED MISSING DATA RATHER THAN FINDING THE SOURCE"
#DF SUMMARIZED BY SUBJECT
df_subject <- df_raw %>% group_by(pid, TASK, DATASET) %>% dplyr::summarise(
  n_utterances = n()
)

#VISUALIZE PARTICIPANTS
gf_bar( ~ TASK, fill = ~DATASET, data = df_raw) %>% 
gf_facet_grid(.~pid)

#VISUALIZE TOTALS
gf_bar (~ TASK, fill = ~DATASET, data = df_raw)

gf_bar (~ DATASET, fill = ~TASK, data = df_raw)

#MOSAIC PLOT
vcd::mosaic(main="Proportion of Utterances by TASK and DATASET",
            data = df_raw, TASK ~ DATASET, rot_labels=c(0,90,0,0), 
            offset_varnames = c(left = 4.5), offset_labels = c(left = -0.5),just_labels = "right",
            spacing = spacing_dimequal(unit(1:2, "lines")))

mosaic(formula = ~DATASET + TASK, 
       data = df_raw,
       main = "Proportion of Utterances by TASK and DATASET", 
       sub = "u = 734 utterance codes",
       labeling = labeling_values,
       labeling_args = list(set_varnames = c(graph = "TASK",
                            datset = "DATASET")))

2.1 UTTERANCES by DATASET

How much variance in number of utterances is explained DATASET, TASK and PARTICIPANT?

2.1.1 OLS Fixed Effects Models

#NUMBER UTTERANCES predicted by DATASET + TASK --> OLS LINEAR REGRESSION
print("OLS-LM, UTTERANCES ~ DATASET + TASK")
## [1] "OLS-LM, UTTERANCES ~ DATASET + TASK"
m1 <- lm(n_utterances ~ DATASET + TASK, data = df_subject)
paste("Model")
## [1] "Model"
summ(m1)
Observations 26
Dependent variable n_utterances
Type OLS linear regression
F(2,23) 1.25
R² 0.10
Adj. R² 0.02
Est. S.E. t val. p
(Intercept) 33.80 4.69 7.21 0.00
DATASETspace -7.90 5.56 -1.42 0.17
TASKixn -3.24 5.56 -0.58 0.57
Standard errors: OLS 
paste("Partition Variance")
## [1] "Partition Variance"
anova(m1)
## Analysis of Variance Table
## 
## Response: n_utterances
##           Df Sum Sq Mean Sq F value Pr(>F)
## DATASET    1  432.2  432.15  2.1614 0.1551
## TASK       1   67.8   67.75  0.3388 0.5662
## Residuals 23 4598.7  199.94
paste("Confidence Interval on Parameter Estimates")
## [1] "Confidence Interval on Parameter Estimates"
confint(m1)
##                  2.5 %    97.5 %
## (Intercept)   24.10554 43.498855
## DATASETspace -19.41210  3.602573
## TASKixn      -14.74543  8.269240
report(m1) #sanity check
## We fitted a linear model (estimated using OLS) to predict n_utterances with
## DATASET and TASK (formula: n_utterances ~ DATASET + TASK). The model explains a
## statistically not significant and weak proportion of variance (R2 = 0.10, F(2,
## 23) = 1.25, p = 0.305, adj. R2 = 0.02). The model's intercept, corresponding to
## DATASET = happiness and TASK = static, is at 33.80 (95% CI [24.11, 43.50],
## t(23) = 7.21, p < .001). Within this model:
## 
##   - The effect of DATASET [space] is statistically non-significant and negative
## (beta = -7.90, 95% CI [-19.41, 3.60], t(23) = -1.42, p = 0.169; Std. beta =
## -0.55, 95% CI [-1.36, 0.25])
##   - The effect of TASK [ixn] is statistically non-significant and negative (beta
## = -3.24, 95% CI [-14.75, 8.27], t(23) = -0.58, p = 0.566; Std. beta = -0.23,
## 95% CI [-1.03, 0.58])
## 
## Standardized parameters were obtained by fitting the model on a standardized
## version of the dataset. 95% Confidence Intervals (CIs) and p-values were
## computed using a Wald t-distribution approximation.
plot_model(m1,  show.intercept = TRUE)

check_model(m1)

#NUMBER UTTERANCES predicted by DATASET X TASK --> LINEAR REGRESSION
print("OLS-LM, UTTERANCES ~ DATASET * TASK")
## [1] "OLS-LM, UTTERANCES ~ DATASET * TASK"
m2 <- lm(n_utterances ~ DATASET * TASK, data = df_subject)
paste("Model")
## [1] "Model"
summ(m2)
Observations 26
Dependent variable n_utterances
Type OLS linear regression
F(3,22) 1.23
R² 0.14
Adj. R² 0.03
Est. S.E. t val. p
(Intercept) 36.57 5.32 6.87 0.00
DATASETspace -13.90 7.84 -1.77 0.09
TASKixn -9.24 7.84 -1.18 0.25
DATASETspace:TASKixn 12.00 11.08 1.08 0.29
Standard errors: OLS 
paste("Partition Variance")
## [1] "Partition Variance"
anova(m2)
## Analysis of Variance Table
## 
## Response: n_utterances
##              Df Sum Sq Mean Sq F value Pr(>F)
## DATASET       1  432.2  432.15  2.1775 0.1542
## TASK          1   67.8   67.75  0.3414 0.5650
## DATASET:TASK  1  232.6  232.62  1.1721 0.2907
## Residuals    22 4366.1  198.46
paste("Confidence Interval on Parameter Estimates")
## [1] "Confidence Interval on Parameter Estimates"
confint(m2)
##                          2.5 %    97.5 %
## (Intercept)           25.52890 47.613954
## DATASETspace         -30.15892  2.349396
## TASKixn              -25.49225  7.016062
## DATASETspace:TASKixn -10.98685 34.986850
report(m2) #sanity check
## We fitted a linear model (estimated using OLS) to predict n_utterances with
## DATASET and TASK (formula: n_utterances ~ DATASET * TASK). The model explains a
## statistically not significant and moderate proportion of variance (R2 = 0.14,
## F(3, 22) = 1.23, p = 0.322, adj. R2 = 0.03). The model's intercept,
## corresponding to DATASET = happiness and TASK = static, is at 36.57 (95% CI
## [25.53, 47.61], t(22) = 6.87, p < .001). Within this model:
## 
##   - The effect of DATASET [space] is statistically non-significant and negative
## (beta = -13.90, 95% CI [-30.16, 2.35], t(22) = -1.77, p = 0.090; Std. beta =
## -0.97, 95% CI [-2.11, 0.16])
##   - The effect of TASK [ixn] is statistically non-significant and negative (beta
## = -9.24, 95% CI [-25.49, 7.02], t(22) = -1.18, p = 0.251; Std. beta = -0.65,
## 95% CI [-1.79, 0.49])
##   - The effect of DATASET [space] × TASK [ixn] is statistically non-significant
## and positive (beta = 12.00, 95% CI [-10.99, 34.99], t(22) = 1.08, p = 0.291;
## Std. beta = 0.84, 95% CI [-0.77, 2.45])
## 
## Standardized parameters were obtained by fitting the model on a standardized
## version of the dataset. 95% Confidence Intervals (CIs) and p-values were
## computed using a Wald t-distribution approximation.
plot_model(m2,  show.intercept = TRUE)

check_model(m2)

2.1.2 POISSON Fixed Effects Models

#NUMBER UTTERANCES predicted by DATASET + TASK --> POISSON DISTRIBUTION
print("GLM-POISSON, UTTERANCES ~ DATASET + TASK")
## [1] "GLM-POISSON, UTTERANCES ~ DATASET + TASK"
p.1 <- glm(n_utterances ~ DATASET + TASK, data = df_subject, family = "poisson")
paste("Model")
## [1] "Model"
summ(p.1)
Observations 26
Dependent variable n_utterances
Type Generalized linear model
Family poisson
Link log
𝛘²(2) 17.76
Pseudo-R² (Cragg-Uhler) 0.50
Pseudo-R² (McFadden) 0.06
AIC 292.63
BIC 296.41
Est. S.E. z val. p
(Intercept) 3.53 0.06 60.36 0.00
DATASETspace -0.28 0.07 -3.77 0.00
TASKixn -0.11 0.07 -1.55 0.12
Standard errors: MLE 
paste("Partition Variance")
## [1] "Partition Variance"
anova(p.1)
## Analysis of Deviance Table
## 
## Model: poisson, link: log
## 
## Response: n_utterances
## 
## Terms added sequentially (first to last)
## 
## 
##         Df Deviance Resid. Df Resid. Dev
## NULL                       25     172.78
## DATASET  1  15.3616        24     157.42
## TASK     1   2.4022        23     155.01
paste("Confidence Interval on Parameter Estimates")
## [1] "Confidence Interval on Parameter Estimates"
confint(p.1)
## Waiting for profiling to be done...
##                   2.5 %      97.5 %
## (Intercept)   3.4103351  3.63942450
## DATASETspace -0.4292558 -0.13583604
## TASKixn      -0.2606450  0.03037217
report(p.1) #sanity check
## We fitted a poisson model (estimated using ML) to predict n_utterances with
## DATASET and TASK (formula: n_utterances ~ DATASET + TASK). The model's
## explanatory power is substantial (Nagelkerke's R2 = 0.50). The model's
## intercept, corresponding to DATASET = happiness and TASK = static, is at 3.53
## (95% CI [3.41, 3.64], p < .001). Within this model:
## 
##   - The effect of DATASET [space] is statistically significant and negative (beta
## = -0.28, 95% CI [-0.43, -0.14], p < .001; Std. beta = -0.28, 95% CI [-0.43,
## -0.14])
##   - The effect of TASK [ixn] is statistically non-significant and negative (beta
## = -0.11, 95% CI [-0.26, 0.03], p = 0.122; Std. beta = -0.11, 95% CI [-0.26,
## 0.03])
## 
## Standardized parameters were obtained by fitting the model on a standardized
## version of the dataset. 95% Confidence Intervals (CIs) and p-values were
## computed using a Wald z-distribution approximation.
plot_model(p.1,show.intercept = TRUE)

check_model(p.1)

#NUMBER UTTERANCES predicted by DATASET * TASK --> POISSON DISTRIBUTION
print("GLM-POISSON, UTTERANCES ~ DATASET X TASK")
## [1] "GLM-POISSON, UTTERANCES ~ DATASET X TASK"
p.2 <- glm(n_utterances ~ DATASET * TASK, data = df_subject, family = "poisson")
paste("Model")
## [1] "Model"
summ(p.2)
Observations 26
Dependent variable n_utterances
Type Generalized linear model
Family poisson
Link log
𝛘²(3) 25.01
Pseudo-R² (Cragg-Uhler) 0.62
Pseudo-R² (McFadden) 0.08
AIC 287.38
BIC 292.42
Est. S.E. z val. p
(Intercept) 3.60 0.06 57.59 0.00
DATASETspace -0.48 0.11 -4.51 0.00
TASKixn -0.29 0.10 -2.91 0.00
DATASETspace:TASKixn 0.41 0.15 2.68 0.01
Standard errors: MLE 
paste("Partition Variance")
## [1] "Partition Variance"
anova(p.2)
## Analysis of Deviance Table
## 
## Model: poisson, link: log
## 
## Response: n_utterances
## 
## Terms added sequentially (first to last)
## 
## 
##              Df Deviance Resid. Df Resid. Dev
## NULL                            25     172.78
## DATASET       1  15.3616        24     157.42
## TASK          1   2.4022        23     155.01
## DATASET:TASK  1   7.2491        22     147.77
paste("Confidence Interval on Parameter Estimates")
## [1] "Confidence Interval on Parameter Estimates"
confint(p.2)
## Waiting for profiling to be done...
##                           2.5 %      97.5 %
## (Intercept)           3.4742131  3.71931074
## DATASETspace         -0.6887490 -0.27241989
## TASKixn              -0.4887608 -0.09637294
## DATASETspace:TASKixn  0.1101534  0.70460435
report(p.2) #sanity check
## We fitted a poisson model (estimated using ML) to predict n_utterances with
## DATASET and TASK (formula: n_utterances ~ DATASET * TASK). The model's
## explanatory power is substantial (Nagelkerke's R2 = 0.62). The model's
## intercept, corresponding to DATASET = happiness and TASK = static, is at 3.60
## (95% CI [3.47, 3.72], p < .001). Within this model:
## 
##   - The effect of DATASET [space] is statistically significant and negative (beta
## = -0.48, 95% CI [-0.69, -0.27], p < .001; Std. beta = -0.48, 95% CI [-0.69,
## -0.27])
##   - The effect of TASK [ixn] is statistically significant and negative (beta =
## -0.29, 95% CI [-0.49, -0.10], p = 0.004; Std. beta = -0.29, 95% CI [-0.49,
## -0.10])
##   - The effect of DATASET [space] × TASK [ixn] is statistically significant and
## positive (beta = 0.41, 95% CI [0.11, 0.70], p = 0.007; Std. beta = 0.41, 95% CI
## [0.11, 0.70])
## 
## Standardized parameters were obtained by fitting the model on a standardized
## version of the dataset. 95% Confidence Intervals (CIs) and p-values were
## computed using a Wald z-distribution approximation.
plot_model(p.2,show.intercept = TRUE)

check_model(p.2)

2.1.3 OLS Mixed Effects Models

#NUMBER UTTERANCES predicted by DATASET + TASK | participatnt--> MIXED LINEAR REGRESSION
print("LMER, UTTERANCES ~ DATASET + TASK")
## [1] "LMER, UTTERANCES ~ DATASET + TASK"
mm1 <- lmer(n_utterances ~ DATASET + TASK+ (1|pid), data = df_subject)
paste("Model")
## [1] "Model"
summ(mm1)
Observations 26
Dependent variable n_utterances
Type Mixed effects linear regression
AIC 198.68
BIC 204.97
Pseudo-R² (fixed effects) 0.09
Pseudo-R² (total) 0.65
Fixed Effects
Est. S.E. t val. d.f. p
(Intercept) 33.80 4.17 8.10 20.51 0.00
DATASETspace -7.90 3.39 -2.33 11.00 0.04
TASKixn -3.24 3.39 -0.96 11.00 0.36
p values calculated using Satterthwaite d.f.
Random Effects
Group Parameter Std. Dev.
pid (Intercept) 10.98
Residual 8.61
Grouping Variables
Group # groups ICC
pid 13 0.62 
paste("Partition Variance")
## [1] "Partition Variance"
anova(mm1)
## Type III Analysis of Variance Table with Satterthwaite's method
##         Sum Sq Mean Sq NumDF DenDF F value  Pr(>F)  
## DATASET 403.75  403.75     1    11  5.4421 0.03967 *
## TASK     67.75   67.75     1    11  0.9132 0.35980  
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
paste("Confidence Interval on Parameter Estimates")
## [1] "Confidence Interval on Parameter Estimates"
confint(mm1)
## Computing profile confidence intervals ...
##                   2.5 %    97.5 %
## .sig01         5.590803 17.904130
## .sigma         5.634678 12.309924
## (Intercept)   25.679890 41.924506
## DATASETspace -14.494243 -1.315280
## TASKixn       -9.827577  3.351386
report(mm1) #sanity check
## We fitted a linear mixed model (estimated using REML and nloptwrap optimizer)
## to predict n_utterances with DATASET and TASK (formula: n_utterances ~ DATASET
## + TASK). The model included pid as random effect (formula: ~1 | pid). The
## model's total explanatory power is substantial (conditional R2 = 0.65) and the
## part related to the fixed effects alone (marginal R2) is of 0.09. The model's
## intercept, corresponding to DATASET = happiness and TASK = static, is at 33.80
## (95% CI [25.12, 42.48], t(21) = 8.10, p < .001). Within this model:
## 
##   - The effect of DATASET [space] is statistically significant and negative (beta
## = -7.90, 95% CI [-14.95, -0.86], t(21) = -2.33, p = 0.030; Std. beta = -0.55,
## 95% CI [-1.05, -0.06])
##   - The effect of TASK [ixn] is statistically non-significant and negative (beta
## = -3.24, 95% CI [-10.28, 3.81], t(21) = -0.96, p = 0.350; Std. beta = -0.23,
## 95% CI [-0.72, 0.27])
## 
## Standardized parameters were obtained by fitting the model on a standardized
## version of the dataset. 95% Confidence Intervals (CIs) and p-values were
## computed using a Wald t-distribution approximation.
plot_model(mm1,  show.intercept = TRUE)

check_model(mm1)

#NUMBER UTTERANCES predicted by DATASET * TASK  | participatnt--> MIXED LINEAR REGRESSION
print("LMER, UTTERANCES ~ DATASET X TASK")
## [1] "LMER, UTTERANCES ~ DATASET X TASK"
mm2 <- lmer(n_utterances ~ DATASET * TASK + (1|pid), data = df_subject)
paste("Model")
## [1] "Model"
summ(mm2)
Observations 26
Dependent variable n_utterances
Type Mixed effects linear regression
AIC 192.83
BIC 200.38
Pseudo-R² (fixed effects) 0.13
Pseudo-R² (total) 0.67
Fixed Effects
Est. S.E. t val. d.f. p
(Intercept) 36.57 5.32 6.87 15.80 0.00
DATASETspace -13.90 7.84 -1.77 15.80 0.10
TASKixn -9.24 7.84 -1.18 15.80 0.26
DATASETspace:TASKixn 12.00 14.13 0.85 11.00 0.41
p values calculated using Satterthwaite d.f.
Random Effects
Group Parameter Std. Dev.
pid (Intercept) 11.15
Residual 8.61
Grouping Variables
Group # groups ICC
pid 13 0.63 
paste("Partition Variance")
## [1] "Partition Variance"
anova(mm2)
## Type III Analysis of Variance Table with Satterthwaite's method
##              Sum Sq Mean Sq NumDF DenDF F value  Pr(>F)  
## DATASET      403.75  403.75     1    11  5.4421 0.03967 *
## TASK          67.75   67.75     1    11  0.9132 0.35980  
## DATASET:TASK  53.48   53.48     1    11  0.7208 0.41398  
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
paste("Confidence Interval on Parameter Estimates")
## [1] "Confidence Interval on Parameter Estimates"
confint(mm2)
## Computing profile confidence intervals ...
##                           2.5 %    97.5 %
## .sig01                 5.088362 17.291859
## .sigma                 5.634705 12.310118
## (Intercept)           26.436694 46.706164
## DATASETspace         -28.822687  1.013164
## TASKixn              -24.156020  5.679831
## DATASETspace:TASKixn -15.486866 39.486867
report(mm2) #sanity check
## We fitted a linear mixed model (estimated using REML and nloptwrap optimizer)
## to predict n_utterances with DATASET and TASK (formula: n_utterances ~ DATASET
## * TASK). The model included pid as random effect (formula: ~1 | pid). The
## model's total explanatory power is substantial (conditional R2 = 0.67) and the
## part related to the fixed effects alone (marginal R2) is of 0.13. The model's
## intercept, corresponding to DATASET = happiness and TASK = static, is at 36.57
## (95% CI [25.46, 47.68], t(20) = 6.87, p < .001). Within this model:
## 
##   - The effect of DATASET [space] is statistically non-significant and negative
## (beta = -13.90, 95% CI [-30.25, 2.44], t(20) = -1.77, p = 0.091; Std. beta =
## -0.97, 95% CI [-2.12, 0.17])
##   - The effect of TASK [ixn] is statistically non-significant and negative (beta
## = -9.24, 95% CI [-25.59, 7.11], t(20) = -1.18, p = 0.252; Std. beta = -0.65,
## 95% CI [-1.79, 0.50])
##   - The effect of DATASET [space] × TASK [ixn] is statistically non-significant
## and positive (beta = 12.00, 95% CI [-17.48, 41.48], t(20) = 0.85, p = 0.406;
## Std. beta = 0.84, 95% CI [-1.22, 2.90])
## 
## Standardized parameters were obtained by fitting the model on a standardized
## version of the dataset. 95% Confidence Intervals (CIs) and p-values were
## computed using a Wald t-distribution approximation.
plot_model(mm2,  show.intercept = TRUE)

check_model(mm2)

2.1.4 POISSON Mixed Effects Models

#NUMBER UTTERANCES predicted by TASK + DATASET  | participatnt--> POISSON MIXED LINEAR REGRESSION
print("POISSON-MER, UTTERANCES ~ DATASET + TASK")
## [1] "POISSON-MER, UTTERANCES ~ DATASET + TASK"
pmm1 <- glmer(n_utterances ~ TASK + DATASET + (1|pid), data = df_subject, family = "poisson")
paste("Model")
## [1] "Model"
summ(pmm1)
Observations 26
Dependent variable n_utterances
Type Mixed effects generalized linear model
Family poisson
Link log
AIC 206.46
BIC 211.49
Pseudo-R² (fixed effects) 0.10
Pseudo-R² (total) 0.84
Fixed Effects
Est. S.E. z val. p
(Intercept) 3.43 0.13 26.57 0.00
TASKixn -0.09 0.08 -1.19 0.23
DATASETspace -0.28 0.08 -3.65 0.00
Random Effects
Group Parameter Std. Dev.
pid (Intercept) 0.41
Grouping Variables
Group # groups ICC
pid 13 0.15 
paste("Partition Variance")
## [1] "Partition Variance"
anova(pmm1)
## Analysis of Variance Table
##         npar  Sum Sq Mean Sq F value
## TASK       1  3.2811  3.2811  3.2811
## DATASET    1 13.2458 13.2458 13.2458
paste("Confidence Interval on Parameter Estimates")
## [1] "Confidence Interval on Parameter Estimates"
confint(pmm1)
## Computing profile confidence intervals ...
##                   2.5 %      97.5 %
## .sig01        0.2746462  0.66576686
## (Intercept)   3.1540729  3.69253649
## TASKixn      -0.2383290  0.05831842
## DATASETspace -0.4245031 -0.12764249
report(pmm1) #sanity check
## We fitted a poisson mixed model (estimated using ML and Nelder-Mead optimizer)
## to predict n_utterances with TASK and DATASET (formula: n_utterances ~ TASK +
## DATASET). The model included pid as random effect (formula: ~1 | pid). The
## model's total explanatory power is substantial (conditional R2 = 0.84) and the
## part related to the fixed effects alone (marginal R2) is of 0.10. The model's
## intercept, corresponding to TASK = static and DATASET = happiness, is at 3.43
## (95% CI [3.18, 3.68], p < .001). Within this model:
## 
##   - The effect of TASK [ixn] is statistically non-significant and negative (beta
## = -0.09, 95% CI [-0.24, 0.06], p = 0.232; Std. beta = -0.09, 95% CI [-0.24,
## 0.06])
##   - The effect of DATASET [space] is statistically significant and negative (beta
## = -0.28, 95% CI [-0.42, -0.13], p < .001; Std. beta = -0.28, 95% CI [-0.42,
## -0.13])
## 
## Standardized parameters were obtained by fitting the model on a standardized
## version of the dataset. 95% Confidence Intervals (CIs) and p-values were
## computed using a Wald z-distribution approximation.
plot_model(pmm1,  show.intercept = TRUE)

check_model(pmm1)

#NUMBER UTTERANCES predicted by TASK X DATASET  | participatnt--> POISSON MIXED LINEAR REGRESSION
print("POISSON-MER, UTTERANCES ~ DATASET X TASK")
## [1] "POISSON-MER, UTTERANCES ~ DATASET X TASK"
pmm2 <- glmer(n_utterances ~ TASK * DATASET + (1|pid), data = df_subject, family = "poisson")
paste("Model")
## [1] "Model"
summ(pmm2)
Observations 26
Dependent variable n_utterances
Type Mixed effects generalized linear model
Family poisson
Link log
AIC 208.07
BIC 214.36
Pseudo-R² (fixed effects) 0.12
Pseudo-R² (total) 0.84
Fixed Effects
Est. S.E. z val. p
(Intercept) 3.50 0.17 20.99 0.00
TASKixn -0.24 0.25 -0.97 0.33
DATASETspace -0.43 0.25 -1.71 0.09
TASKixn:DATASETspace 0.30 0.48 0.63 0.53
Random Effects
Group Parameter Std. Dev.
pid (Intercept) 0.40
Grouping Variables
Group # groups ICC
pid 13 0.14 
paste("Partition Variance")
## [1] "Partition Variance"
anova(pmm2)
## Analysis of Variance Table
##              npar  Sum Sq Mean Sq F value
## TASK            1  3.3827  3.3827  3.3827
## DATASET         1 13.2136 13.2136 13.2136
## TASK:DATASET    1  0.4015  0.4015  0.4015
paste("Confidence Interval on Parameter Estimates")
## [1] "Confidence Interval on Parameter Estimates"
confint(pmm2)
## Computing profile confidence intervals ...
##                           2.5 %    97.5 %
## .sig01                0.2676997 0.6545456
## (Intercept)           3.1401510 3.8418542
## TASKixn              -0.7570693 0.2834377
## DATASETspace         -0.9484807 0.1001263
## TASKixn:DATASETspace -0.7157996 1.3097794
report(pmm2) #sanity check
## We fitted a poisson mixed model (estimated using ML and Nelder-Mead optimizer)
## to predict n_utterances with TASK and DATASET (formula: n_utterances ~ TASK *
## DATASET). The model included pid as random effect (formula: ~1 | pid). The
## model's total explanatory power is substantial (conditional R2 = 0.84) and the
## part related to the fixed effects alone (marginal R2) is of 0.12. The model's
## intercept, corresponding to TASK = static and DATASET = happiness, is at 3.50
## (95% CI [3.17, 3.82], p < .001). Within this model:
## 
##   - The effect of TASK [ixn] is statistically non-significant and negative (beta
## = -0.24, 95% CI [-0.72, 0.25], p = 0.333; Std. beta = -0.24, 95% CI [-0.72,
## 0.25])
##   - The effect of DATASET [space] is statistically non-significant and negative
## (beta = -0.43, 95% CI [-0.92, 0.06], p = 0.088; Std. beta = -0.43, 95% CI
## [-0.92, 0.06])
##   - The effect of TASK [ixn] × DATASET [space] is statistically non-significant
## and positive (beta = 0.30, 95% CI [-0.63, 1.24], p = 0.526; Std. beta = 0.30,
## 95% CI [-0.63, 1.24])
## 
## Standardized parameters were obtained by fitting the model on a standardized
## version of the dataset. 95% Confidence Intervals (CIs) and p-values were
## computed using a Wald z-distribution approximation.
plot_model(pmm2,  show.intercept = TRUE)

check_model(pmm2)